The suitability of iPhone recordings for the acoustic measures of speech and voice quality

摘要

This study examined the quality of iPhone recordings for acoustic measurements of speech and voice quality. A selection of acoustic measures were extracted from voice samples recorded using the “voice memo” application in an iPhone and compared with those derived from signals directly digitized (DD) in a laptop via a 12-bit A/D converter. Participants were 11 healthy adults, including six females and five males, aged between 27 to 67 years (Mean = 41.8 years, SD = 16.7). The participant was asked to read the first six sentences of the “rainbow passage”. In addition, two participants were asked to produce sustained vowels (/i/, /a/, and /u/) and a sentence (“We saw two cars”) ten times. The simultaneously recorded iPhone and DD signals were analysed to derive 10 acoustic measures, including spectral tilt for the whole sentence and fundamental frequency (F0), percent jitter, percent shimmer, signal-to-noise ratio, amplitude of the first harmonic relative to that of the second harmonic, singing power ratio, and frequencies of the first and second formants (F1 and F2), and vowel space area for the vowel segment. A series of Pearson’s correlation procedures revealed that measures from iPhone and DD signals were highly correlated. Findings of the vowel effect on the experimental measures obtained from iPhone signals were consistent with those from DD signals. However, the mean normalized absolute differences between measures from iPhone and DD signals are optimal (i.e., lower than 20%) only for F0, F1, and F2. These findings suggest that iPhone recordings are as adequate as other types of high quality digital recordings for acoustic measurements of voice quality but most voice measures from different digital recording systems are not directly comparable.